PostgreSQL

PostgreSQL is the database for Vercel and for any deployment that needs a managed, shared store. The adapter reuses the same portable SQL repository as D1, translated to the PostgreSQL dialect, so behaviour is identical across backends.

No driver is bundled. You pass a client, which keeps connection pooling and credentials in your hands.

import { lazyStatusDatabase } from "@heyo-sh/heyo-status/db/lazy";import { postgres } from "@heyo-sh/heyo-status/deployments/vercel";import { drizzle } from "drizzle-orm/node-postgres";import { Pool } from "pg";export const database = lazyStatusDatabase(() => {  const connectionString = process.env.DATABASE_URL;  if (!connectionString) {    throw new Error("DATABASE_URL is required when using the PostgreSQL database overlay.");  }  return postgres(drizzle(new Pool({ connectionString })));});

Accepted clients

postgres() takes anything with a compatible query surface:

import { postgres } from "@heyo-sh/heyo-status/db/postgres";import { Pool } from "pg";const database = postgres(new Pool({ connectionString: process.env.DATABASE_URL }));

Both import paths are the same implementation: @heyo-sh/heyo-status/db/postgres and @heyo-sh/heyo-status/deployments/vercel.

Migrations

heyo-status db migrate

Through the lazy wrapper this calls PostgresStatusDatabase.migrate(), which applies the bundled migrations and records them in __heyo_status_migrations. Run it:

• once per environment before the first deploy, and
• after upgrading the package, as part of your release job.

// A one-off deployment migration job.import { database } from "./heyo-status.database";await database.migrate();

Request handlers never migrate automatically, so a cold start cannot race schema changes.

Options

postgres(client);postgres(client, { migrate: true });

migrate: true applies migrations on construction — appropriate for local and test databases, not for a shared production one.

Connection pooling

Serverless functions and pooled PostgreSQL need care:

• On Vercel, prefer a pooled connection string (PgBouncer, Neon pooler, Supabase pooler). A direct connection per invocation exhausts the server.
• Keep the pool small; a status page issues a handful of queries per request.
• A status page is read-heavy, and cache.snapshotTtlMs removes most repeated reads inside one instance.

Operational notes

• Timestamps are stored as epoch milliseconds, exactly as in SQLite and D1, so history is portable between backends.
• A unique index on the subscriber destination is what makes concurrent subscription requests resolve to one row.
• heyo-status check run and db seed work against PostgreSQL from a developer machine as long as DATABASE_URL is reachable.